AI · B2B SaaS · MarTech · 2026
B2B marketing teams were stitching demand generation together across separate tools for ICP definition, account discovery, intent analysis, campaign activation and reporting. Lucy collapses all five into one AI-driven platform with Salesforce sync. I scoped the product and owned the full delivery lifecycle — functional scope, flows, architecture and the twelve-week plan that shipped it.
Executive Summary
Demand generation runs on intent signals — who is researching your category right now, and how strongly. The tooling for it had fragmented: one product to define an ideal customer profile, another to discover accounts, another to read intent, another to activate a campaign, another to report on it.
Every one of those handoffs is a place where context is lost and a signal ages. Lucy was scoped to remove the handoffs rather than improve any individual step — a single system covering ICP identification, account discovery, intent analysis, campaign activation and reporting, with Salesforce sync so it lands inside the sales workflow the team already runs.
I owned the full lifecycle: functional scope split between client and admin surfaces, the six-step client flow, the technical architecture, and a twelve-week delivery plan across six roles. Delivery ran on two-week sprints with a collaboratively prioritised backlog rather than fixed milestones, because a platform whose intent logic keeps evolving cannot be specified once and then built to a frozen scope.
Section 02 — The Problem
When ICP lives in one system and discovery in another, nobody can be certain the accounts being scored are the accounts the profile actually described. The reconciliation is manual and it is nobody's job.
Intent decays. A buying signal that reaches a rep a week late produces an outreach that reads as random, which costs more credibility than staying silent would have.
Intelligence that does not reach Salesforce does not reach the person making the call. Any platform that stops at its own dashboard has solved the analysis and skipped the point.
Section 03 — Research & Discovery
The scoping question was which capabilities had to be in one system for the seams to actually disappear. Five came back as non-negotiable — ICP identification, account discovery, intent analysis, campaign activation and reporting — because a gap in any of them re-introduces a manual handoff and undoes the premise.
Salesforce integration was pulled into the same phase rather than deferred, for the same reason. Shipping the intelligence first and the CRM sync later would have meant launching a product that stopped one step short of the person who acts on it.
Onboarding was designed as guided configuration with AI-suggested profiles the client reviews and refines — not an automatic profile they inherit. An ICP nobody agreed to is one nobody trusts the scoring from.
Supporting one CRM properly beats supporting three partially, because a half-working sync is worse than none: it produces records the sales team stops trusting. HubSpot and Pipedrive went on the roadmap in writing so the limit read as sequencing rather than an oversight.
Section 04 — Current State Analysis
Writing the future column down at proposal stage was deliberate. Multi-CRM support and custom models are the two things a buyer in this category always asks about, and the difference between a gap and a roadmap is whether you named it before they did.
Before / After
The capabilities are the same either way. What changes is how many times a list of accounts has to leave one system and be reconciled into another before anyone can act on it.
Before
5 tools
Define the ICP in one tool
Discover accounts in another — and reconcile them against the profile by hand
Manual handoffRead intent signals in a third
Move the prioritised list into a campaign tool
Manual handoffReport on performance somewhere else again
Every boundary is a place where context is lost, a signal ages, and no single system holds the whole picture.
After
1 platform
ICP set up or AI-suggested, confirmed by the client
Accounts and contacts discovered and enriched in place
Intent scored and prioritised with source and recency shown
Campaign activated or leads exported — same system
Performance reported back against the same ICP
Salesforce sync carries accounts, contacts and intent scores into the CRM, so the output lands where sales already works.
Section 05 — Competitive Research
In a category this crowded, the realistic alternative is rarely a single rival platform — it is the collection of tools a team has already bought, integrated badly, and trained people on. That is a genuinely strong competitor, because switching costs are paid in retraining rather than licence fees.
This shaped two decisions directly. Salesforce sync went into the first phase so Lucy joined the existing workflow rather than asking teams to abandon it, and consolidation became the product's argument — the pitch is one fewer handoff, not a better intent score than any individual incumbent.
Any incumbent that does one of these five things does it well, and out-scoring a specialist at its own job is a race with no finish line. The defensible claim is the one none of them can make without becoming a different product: four fewer handoffs.
The comparison a buyer actually runs is total cost of the tools being replaced plus the labour of reconciling them. That framing matters more than per-seat pricing, and it only works if the platform genuinely covers all five capabilities.
Section 06 — Key Insights
An intent score on its own is a number a rep has to take on faith. Shown with where the signal came from and how fresh it is, it becomes evidence they can act on and defend.
Beating any single incumbent on its own dimension is a losing race. Removing four handoffs is a claim none of them can match without becoming a different product.
Intent scoring changes as real data arrives. Two-week sprints against a collaboratively prioritised backlog match how the product actually develops; a frozen scope would have locked in assumptions made before any real signal existed, then treated every correction as a change request.
Section 07 — Design Strategy
A gap in any of the five capabilities re-introduces the handoff the product exists to remove.
Salesforce sync shipped in phase one because intelligence that stops at a dashboard never reaches the rep.
Scoring is only trusted downstream if somebody agreed to the profile it was derived from.
Multi-CRM and custom models stated as roadmap read as sequencing; discovered later they read as gaps.
Section 08 — The System
Onboarding and ICP setup, account and contact discovery, intent visibility, campaign activation and export, performance reporting.
Client and campaign management, ICP and intent configuration controls, Salesforce sync monitoring, system health and activity logs.
Salesforce sync for accounts, contacts and intent scores, under role-based permissions with rate-limited APIs.
Section 09 — The Features
Guided onboarding with AI-suggested ICP refinement, anonymous account discovery via website tracking, contact enrichment filtered by role and firmographics, and intent scores shown with their source and recency rather than as a bare number.
In-market buyer counts, average contacts per account and geography, alongside a sample of the accounts the ICP actually matched — shown before the client commits to activation.
Add-to-campaign and lead export workflows with performance reporting, so a signal becomes an action inside the same system that found it.
A lead brief carries the response probability with the signals behind it — funding events, hiring, podcast mentions — plus what to say and what to avoid, then exports to Salesforce, Excel or PDF.
Secure admin authentication, client and campaign management, ICP and intent configuration, Salesforce sync monitoring, and system health and activity logs.
Section 10 — User Flow
Supabase Auth with role-based permissions; the admin surface is a separate authenticated context.
Guided configuration. The profile is proposed by the system and confirmed by the client, never assumed.
Anonymous account discovery from website tracking, then contact enrichment filtered by role and firmographics.
Each score carries its source and recency, so the ranking can be interrogated rather than trusted blindly.
The activation step lives in the same system as the discovery, which is the entire argument for consolidation.
Engagement reporting closes the loop back to the ICP, so the profile improves with evidence.
The admin flow runs alongside it in four steps: secure login, review of client configurations, monitoring of campaigns and integrations, then system settings and reporting.
The point of listing it this way is that nothing in the sequence leaves the platform. On the previous stack, three of those six steps ended with an export and a re-import somewhere else.
Section 11 — End-User Experience
Nobody logs into a demand-gen platform to admire a scoring model. They log in to find out where to spend the day. Every design decision was measured against how quickly someone gets from opening the product to a prioritised, defensible list they can act on.
Showing source and recency alongside each score is the part that carries this. It converts a ranking into an argument the user can make to their own sales team, which is what determines whether the tool gets used after the first week.
The behavioural test for a product like this is whether reps work the list top-down. If they re-sort it by their own instinct, the scoring has not earned trust yet — and no amount of model accuracy fixes that if the reasoning is not visible alongside the score.
Section 12 — Impact & Outcomes
Delivery was deliberately structured around a collaboratively prioritised backlog with weekly progress updates and sprint demos — so the intent logic could keep changing after launch without a change-request conversation every time.
Post-launch figures are not published, so none are claimed. The measure that would settle it is intent scoring against actual conversions — everything else describes activity, and only that comparison tells you whether the prioritisation was right.
Section 13 — Reflection
The scoring engine had a two-week slot in the plan, which assumes the hard part is implementation. The hard part is agreeing what correct means — and that is a conversation better had before the sprint than during it.
Replacing five specialised tools with one platform means every capability is now compared against a product that did nothing else. The bet is that removing the handoffs outweighs being second-best at any single step — and that is validated by retention, not by launch.